Romance Of Hr And Performance

http://hum.sagepub.com

Human Relations

DOI: 10.1177/0018726705055032 2005; 58; 429 Human Relations

Toby D. Wall and Stephen J. Wood and the case for big science

The romance of human resource management and business performance,

http://hum.sagepub.com/cgi/content/abstract/58/4/429 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

On behalf of:

The Tavistock Institute

can be found at:Human Relations Additional services and information for

http://hum.sagepub.com/cgi/alerts Email Alerts:

http://hum.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

http://hum.sagepub.com/cgi/content/refs/58/4/429SAGE Journals Online and HighWire Press platforms):

(this article cites 36 articles hosted on the Citations

© 2005 The Tavistock Institute. All rights reserved. Not for commercial use or unauthorized distribution. at Middlesex University on February 22, 2008 http://hum.sagepub.comDownloaded from

http://www.tavinstitute.org/index.php

http://hum.sagepub.com/cgi/alerts

http://hum.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.com/journalsPermissions.nav

http://hum.sagepub.com/cgi/content/refs/58/4/429


The romance of human resourcemanagement and business performance,and the case for big scienceToby D. Wall and Stephen J. Wood

A B S T R AC T It is often assumed that research over the last decade has established

an effect of human resource management (HRM) practices on

organizational performance. Our critical assessment of existing

studies finds that, although collectively they have opened up a

promising line of inquiry, their methodological limitations make such

a conclusion premature. We argue that future progress depends on

using stronger research methods and design that, in turn, will require

large-scale long-term research at a level of magnitude that probably

can only be achieved through partnerships between research, prac-

titioner and government communities. We conclude that progress

so far justifies investment in such big science.

K E Y WO R D S company performance � high performance work organization �

high performance work systems � human resourcemanagement � organizational performance

Picture the scene. A leading scholar specializing in human resource manage-ment (HRM) is called to court as an expert witness by an internationalcompany that has brought a case against a firm of consultants. The companyhas paid several hundred thousand pounds in consultancy fees, and investedmany times that amount in its personnel function, to introduce the ‘perform-ance-enhancing’ HRM practices recommended by the consultants. Threeyears on there is no evident return on their investment. The company is suing

4 2 9

Human Relations

DOI: 10.1177/0018726705055032

Volume 58(4): 429–462

Copyright © 2005

The Tavistock Institute ®

SAGE Publications

London, Thousand Oaks CA,

New Delhi

www.sagepublications.com

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 429



the consultants on the grounds that they created misleading expectations ofthe effect of HRM practices on performance. The expert witness is asked toprepare a report addressing two questions: (i) whether it was reasonable ofthe consultants to assume an effect of HRM practices on performance on thebasis of both researchers’ accounts of their own findings and how prac-titioners have been encouraged to view them; and (ii) the extent to which theevidence can, in fact, be interpreted as establishing such a causal link. In thisarticle we present the report that we would have submitted were we theexpert witness. Having adopted this format, we conclude by considering theimplications of our assessment for future research. First, however, we intro-duce the core concepts.

Conceptual background

HRM is a term used to represent that part of an organization’s activitiesconcerned with the recruitment, development and management of itsemployees (Wood & Wall, 2002). Within that domain, current interest isfocused on HRM systems emphasizing all or most of the following practices:sophisticated selection methods, appraisal, training, teamwork, communi-cations, empowerment, performance-related pay and employment security.Collectively, these are deemed to contribute to the skill and knowledge basewithin the organization, and to employees’ willingness to deploy theirlearning to the benefit of the organization. For this reason, authors have usedlabels such as ‘human capital-enhancing’ (Youndt et al., 1996), ‘highcommitment’ (Wood & de Menezes, 1998) or ‘high involvement’ (Guthrie,2001; Vandenberg et al., 1999) to characterize the approach. They oftencontrast this with traditional, Tayloristic or control approaches to manage-ment, which emphasize low skill, limited employee discretion and a tightdivision of labour (e.g. Arthur, 1994; Lawler, 1986; Walton, 1985).

The rationale for focusing on the high-commitment approach to HRM lies in the assumption that the practices will enhance organizationalperformance. Related to this is the view that the practices will be a sourceof unique and sustainable advantage because, unlike other initiatives such asthe introduction of new technology, they result in skills and knowledge thatare largely organization specific and are therefore difficult to imitate (e.g. Barney, 1991, 1995). Companies can copy one another’s technologiesmuch more easily than they can their human resource capabilities.

The link between HRM and performance has been conceived in avariety of ways. The simplest view is that practices are additive (the morethe better, e.g. Guest & Hoque, 1994), and that they enhance performanceregardless of circumstance (a universal effect, e.g. Pfeffer, 1994). Alternative

Human Relations 58(4)4 3 0

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 430



perspectives emphasize various kinds of ‘fit’, of which three main types havebeen identified (Wood, 1999). ‘Internal fit’ posits synergy among the prac-tices, meaning that their collective effect will be greater than the sum of theirindividual parts. For example, it might be argued that selecting able peoplewithout training them, or training employees but not empowering them touse that training, will have little effect; whereas implementing the three prac-tices together will. Indeed, Barney (1995) goes so far as to argue that indi-vidual practices ‘have limited ability to generate competitive advantage inisolation’ (p. 56).

The second type is ‘organizational fit’, which concerns the role of HRMin enhancing the effectiveness of other organizational practices or technolo-gies, and vice versa. Lawler et al. (1995), for example, link HRM to totalquality management (TQM), arguing that the two sets of practices are‘complementary in their impact on organizational performance’ (p. 144).Similarly, MacDuffie (1995) sees high commitment HRM practices asintegral to the effectiveness of lean production (Womack et al., 1990) initia-tives and vice versa.

The third and final type is ‘strategic fit’. This assumes that HRM prac-tices need to be aligned with the organization’s strategy (e.g. in the case ofprivate businesses to their competitive strategy) to have their full effect onperformance (e.g. Miles & Snow, 1984; Schuler & Jackson, 1987; Youndtet al., 1996).

Empirical investigation of fit typically involves testing for interactioneffects (e.g. among different practices, or between HRM system measuresand strategy). Because fit is central to many accounts of the effects of HRMon performance, it is relevant to determine the extent to which such inter-actions have been examined. In practice, as we shall see, much empiricalwork ignores or pays only limited attention to issues of fit, consciously or bydefault taking the universal thesis propounded by Pfeffer (1994) and others.

Is an HRM effect on performance presumed by researchersand practitioners?

We can now return to the first question we have been asked to consider asexpert witnesses, whether authors portray research findings as confirming acausal effect of HRM on organizational performance, in support of theguiding theory. There can be little doubt that such an interpretation is wide-spread. Early reviews of the field fuelled this perception as they concludedthat studies supported the universal thesis (e.g. Becker & Huselid, 1998;Guest, 1997). Becker and Gerhart (1996: 779), for example, argue that‘conceptual and empirical work . . . has progressed enough to suggest that

Wall & Wood The romance of HRM 4 3 1

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 431



the role of human resources can be crucial’. Such views have permeatedsubsequent studies. Way (2002: 765), for instance, states that ‘Theoreticaland empirical HRM research has led to a general consensus that the methodused by a firm to manage its workforce can have a positive impact on firmperformance’.

A more subtle way in which authors imply an established performanceeffect lies in the labels many use to denote salient HRM systems. Forexample, in his seminal study, Huselid (1995) uses the expression ‘highperformance work practices’ (p. 635), mirroring Lawler et al.’s (1995, 1998)choice of ‘high performance organizations’ for their book titles. Subse-quently, Delaney and Huselid (1996) refer to ‘performance-enhancing’(p. 949) HRM practices, Kalleberg and Moody (1996) to ‘high performancework organizations’ (p. 114), and Appelbaum et al. (2000) and Way (2002:765) to ‘high performance work systems’. The titles of many articles carrythe same message. For example, an early article by Arthur (1994) is entitled‘Effects of human resource systems on manufacturing performance andturnover’. Following Huselid’s (1995) lead in his article on ‘The impact ofhuman resource management practices on turnover, productivity and corpo-rate financial performance’ (our italics), many other authors, also describingcross-sectional results, use equivalent causal terminology (e.g. Ahmad &Schroeder, 2003; Bae & Lawler, 2000; Delaney & Huselid, 1996; Jayaramet al., 1999; Vandenberg et al., 1999; Wright et al., 2003). The idea of causal-ity is thus embedded within the choice of label for HRM systems and in the wording of titles. Such terminology is in danger of appearing to prejudgethe very relationship under investigation, as if the researcher’s role were tofind evidence for a widely expected (and one might say, even hoped for)relationship.

A further way in which the discourse of research implies an establishedeffect of HRM on performance is through the argument that it is now timeto move on to other issues. In this vein, Delery (1998) suggests that ‘Estab-lishing that HRM practices are linked with firm effectiveness is an import-ant first step . . . however, there is little understanding of the mechanismsthrough which HRM practices influence effectiveness’ (p. 289). Similarly,there have been calls to broaden the outcomes studied to cover those of moreimmediate interest to employees. Osterman’s (2000) study of ‘High perform-ance work organization’ practices follows this lead, as he set his objective todetermine ‘whether their productivity and quality gains redounded toemployees’ benefit’ (p. 179). He examines the not unimportant question ofwhether these practices are associated with layoff rates and compensationgains; but in so doing he simply assumes, without proof, that there are gainsto redound (i.e. share among employees).


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 432



A final indication of the way in which a causal effect is assumed relatesto practice. In their review of research, Becker and Gerhart (1996) set theirgoal as being to ‘demonstrate to senior human resources (HR) and linemanagers that their HR systems represent a largely untapped opportunity toimprove firm performance’ (p. 780). That message has clearly got through,as evident from Cooper’s (2000) article in a leading UK HRM practitionerjournal, which concludes: ‘Academic studies have established a link betweenHR and profitability. The race is now on to find out how and why peoplepolicies make a difference to shareholders’ (p. 28). Again, Philpott (2002)concludes that:

What is missing, says the CIPD [the UK’s Chartered Institute of Person-nel Development], is a genuine appreciation both that people manage-ment holds the key to increased productivity and that meeting theobjective requires the appropriate application of a range of peoplemanagement practices. In general terms these practices encompass,recruitment, training, job appraisal and reward, job design, job qualityand communication with staff.

(p. 5)

One should record that some have been more circumspect in their assess-ments (e.g. Delaney & Godard, 2001; Marchington & Grugulis, 2000;Wood, 1999). Wright and Gardner (2003), for example, go only so far as tosuggest that evidence mounts that HR practices are at least ‘weakly relatedto firm performance’ (p. 312). Godard (2004) is even more sceptical. Thoughnot doubting that HRM ‘can be highly effective in some work places’(p. 355), he concludes that, especially in liberal market economies, the gener-alizability is likely to be low. Nevertheless, the message conveyed by a largepart of the literature is that HRM practices do promote performance.

Expert witness testimony: part 1

Our evidence on the first question is clearly in favour of the defence. It showsthat it was reasonable for the consultants to assume a positive effect of HRMpractices on performance on the basis both of researchers’ portrayal of theirown findings and how these have been presented in non-academic circles.

Does the evidence support an HRM–performance effect?

To address the second question, concerning the extent to which the evidencecan be interpreted as establishing a causal effect of HRM on performance,


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 433



we evaluate key empirical studies. What follows is an account of the methodwe adopted, and our findings.

Sampling the studies

We used five criteria to select the studies to be evaluated. First, we restrictedour choice to those appearing in reputable refereed journals, to ensure thatwe were assessing high-quality studies. This meant that other influentialwork, published, for example, in books or reports, was excluded (e.g. Appel-baum et al., 2000; Ichniowski, 1990; Lawler et al., 1995, 1998). Second, werestricted our selection to studies published from 1994 onwards, whenresearch on HRM and performance first came to prominence. Third, weincluded only those studies covering multiple HRM practices, because thefocus is on whether the HRM system as a whole promotes overall organiz-ational performance. This excluded any studies of single practices such astraining, sophisticated selection, or job design. Fourth, we included onlystudies using measures of economic performance such as productivity, profitor return on assets. Thus we excluded studies limited to other performanceindicators such as labour turnover, absence or employee well-being. Finally,because of the detailed assessment involved, we restricted the sample tohighly cited milestone studies of the mid- to late-1990s, and a selection ofmore recent ones (whose citation rate is yet to be determined). The studieswe used are shown in Table 1.

Assessing the studies

In order to assess the evidence, we identified key criteria relevant to gener-alization and causal interpretation against which to judge studies individu-ally and collectively. Those criteria concern: the sample and response rate;the reliability and validity of the HRM measure and source of data for it;the adequacy of the research design; the extent to which other factors havebeen controlled; the strength of the findings on the HRM–performance link;whether there has been a test for fit; and finally if the effects of individualHRM practices have been considered alongside those of the composite HRMmeasure. We expand on these criteria and the rationale for their choice inthe course of presenting findings.

Sample and response rate

The size of samples used affects the power to detect main effects, and evenmore so theoretically predicted synergies (interactions). Coupled with the


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 434



response rate, size also affects the extent to which one can generalize. It canbe seen (Table 1, column 2) that the studies have diverse but often rathersmall (< 300 in 18 of the 25 studies, and < 100 in 9 studies) samples (column2). At the same time, the response rates, though varying between 4 and 84percent, are generally low – albeit better than the 6–28 percent (mean 17.4%)rate reported in an earlier review by Becker and Huselid (1998). This is nota very secure foundation from which to generalize.

HRM dimensions

Evidence that the studies use appropriate and broadly equivalent measuresof HRM is important to justify treating them collectively. As previouscommentators have noted (e.g. Becker & Gerhart, 1996; Dyer & Reeves,1995; Wright & Gardner, 2003) there is diversity across studies in theparticular practices covered. Nonetheless, there is much commonality asstudies typically cover a substantial range of the following: sophisticatedselection, appraisal, training, teamwork, communication, job design,empowerment, participation, performance-related pay/promotion, harmo-nization, and employment security (Table 1, column 3).

HRM measure type and source

In any empirical investigation reliable measurement of independent variables(in this case HRM practices) is of paramount importance; and in cross-sectional research, ensuring such measures are uncontaminated by those ofthe dependent variables (i.e. performance) is of particular significance. Instudies of HRM, the measurement of relevant practices cannot be a straight-forward process, because the focal construct is not of that nature. Whetherthere is a bonus scheme in operation may be easy to establish, but judge-ments of the extent of teamwork or empowerment are much more compli-cated to gauge. Thus, quantifying HRM practices involves judgement.Consequently, it is open to random measurement error. At the same time, itis vulnerable to correlated error in the form of rater bias. For example, moreoptimistic or organizationally committed individuals rating both the prac-tices and performance in their own company might give systematically higherscores than their more pessimistic counterparts elsewhere. Such ‘commonmethod variance’ alone could create spurious relationships among HRMpractices and between them and performance.

Given these considerations, the measurement of HRM practices wouldideally involve: assessments from two or more persons (to determinereliability), the use of the same raters across different organizations (to


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 435



Hu

man

Rela

tion

s 58

(4)

43

6

Table 1 Measurement features of empirical studies on HRM and organizational performance

Study Sample and response HRM dimensions and illustrative HRM measure type and source Dependent variable measures rate (%) components

1 Arthur (1994) 30 US mini steel Two contrasting types of HRM system: Questionnaire; single-source; Self-report: productivity and scrap mills; 56 percenta control and commitment (low and high HR managers rate; same respondent as for

respectively on decentralization, independent variables; for previous participation, training and skill) year

2 Guest and 119 UK Four a priori types: 2 � 2 – whether or Questionnaire; single-source; Self-report: productivity and Hoque (1994) greenfield mainly not claim HRM strategy and use more or principal HR manager or quality; same respondent as for

manufacturing less than half of 22 HRM practices senior line manager independent variables; concurrentcompanies; (e.g. selection, job design)39 percenta

3 Huselid 968 US Two scales: skills and structures Questionnaire; single-source; Objective: productivity, Tobin’s (1995) companies with (e.g. communications, QWL, training, mailed to senior HRM Q and GRATE for subsequent year

100+ staff; grievance procedures); and motivation professionalheterogenous sample; (e.g. performance appraisals, promotion 28 percent on merit)

4 MacDuffie 62 car-assembly Two scales: work systems (participation, Questionnaire; ‘a contact Objective: productivity (labour (1995) plants, worldwide; teams, quality role) and HRM policies person, often the plant hours per vehicle) – adjusted for

69 percenta (selection, performance-related pay, manager’; sections completed vehicle size and number of welds;training). Also a use of buffers (lean by different people; later quality from independent market production) scale visits/telephone checks report; concurrent

5 Delaney and 590 for-profit and Five scales: staffing selectivity, training, Telephone survey; single-source Self-report: organizational and Huselid (1996) non-profit US firms, incentive pay, decentralization, internal (multiple respondents in a few market performance; same

heterogenous; promotion cases); unspecified respondent as for independent 51 percent ‘representatives’ variables; concurrent

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 436

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U

niversity on February 22, 2008


Dow

nloaded from


Wall &

Wood

Th

e ro

man

ce o

f HR

M4

37

6 Delery and 216 US banks Seven scales: internal promotion, Questionnaire; single-source; Objective: return on assets (ROA),Doty (1996) (down to 101 in training, appraisal, profit-sharing, senior HR manager (+ strategy return on equity (ROE);

some analyses); security, participation, job specification; from president) concurrent18 percenta and two strategy measures

7 Koch and 319 US business Separate HR planning index, Hiring Questionnaire; single-source; Objective: labour productivity;McGrath units; 4 percenta index and Development index; and sent to executives concurrent(1996) composite HR sophistication index

(previous three added together) 8 Youndt et al. 97 US manufacturing Two scales: administrative HR Questionnaire; multiple source Self-report: customer alignment

(1996) plants (metal-working (e.g. appraisal, incentives) and human (at least two respondents per (quality), productivity and machine industry); 19 percenta capital-enhancing HR (e.g. selection and plant, mean score); general and efficiency; same respondent as for

training for problem-solving, salaried pay) functional managers independent variables; average of concurrent and subsequent

9 Huselid et al. 293 US firms: Two scales: strategic HRM (teamwork, Questionnaire; single-source; Objective: productivity, GRATE (1997) heterogenous empowerment) and technical HRM executives in HR (92%) and Tobin’s Q; concurrent and

(e.g. manufacturing, (recruitment, training). Ratings are not and line (8%) positions subsequent year finance, misc.); of reported use, but of perceived (effectiveness of HR practices,response rate unclear effectiveness not use per se)

10 Ichniowski 36 steel production Four types of line ranging from innovative Interviews; multiple Objective: productivity (line et al. (1997) lines in 17 companies; HRM (high on incentive pay, selection, respondents; labour relations and uptime); panel data

60 percenta teamwork, employment security, training) operations managers, line to traditional HRM (low on all above workers and trade union repscomponents)

11 Wood and Representative Four types of workplace, ranging from high Interviews; single-source; HR Self-report: productivity,de Menezes sample of 1693 to low HCM (high commitment manager or senior manager productivity change over last 3 (1998) UK workplaces management) – (high HCM showing more responsible for HR years; and financial performance;

(806–926 in analyses); appraisal, information disclosure, same respondent as for 84 percent communication, monthly pay) independent variables; concurrent

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 437

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Hu

man

Rela

tion

s 58

(4)

43

8

Table 1 continued


12 Hoque (1999) 209 UK hotels; Overall HRM (21 practices used, Questionnaire; single-source; Self-report: productivity, service 35 percent including harmonization, job design, respondents unclear quality and financial performance;

training, merit pay) same respondent as for independent variables; concurrent

13 Jayaram et al. 57 firms supplying Five scales: cost, quality, flexibility and Questionnaire; single-source; Self-report: cost, quality, flexibility (1999) automotive industry; time HRM (commitment to and training, CEOs and time performance; same

39 percent goals and support for), plus a generic respondent as for independent scale (broad jobs, autonomy) variables; concurrent

14 Vandenberg 49 North American Five scales: flexibility, incentive Questionnaire; single-source; Objective: return on equity (ROE);et al. (1999) life insurance practices, direction sharing, training, head of HR for subsequent year

companies; response work design rate unclear

15 Wright 38 US petrochemical Four scales: selection, training, pay and Questionnaire; single-source; Self-report: financial performance;et al. (1999) refineries; 20 percent appraisal; with participation as a HR manager independent source (refinery or

contingency variable operational manager); concurrent16 Bae and 138 firms in Korea; Single ‘high involvement HRM strategy’ Questionnaire; single-source; Self-report global measure; same

Lawler (2000) response rate unclear index (five sub-scales covering training, head of HR respondent as for independent empowerment, selection, performance- variables; concurrent. Objective related pay, broad job design) (ROIC) for sub-sample (n = 68) –

concurrent and subsequent year

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 438

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Wall &

Wood

Th

e ro

man

ce o

f HR

M4

39

17 Fey et al. (2000) 101 foreign-owned Eight single-item measures for managers and Questionnaire; single-source; Self-report; overall performance firms in Russia; non-managers separately: promotion on 38 percent HRM manager, (poor to outstanding); same heterogenous sample; merit; job security; technical training; 63 senior managers respondent as for independent 25 percent non-technical training; career planning; variables; concurrent

decentralization; internal recruitment;complaint resolution

18 Cappelli and 433–666 (see Extent of job rotation, self-managed Telephone survey; single- Self-reportb: sales per employee,Neumark research design) teams, teamwork training, cross-training, source; plant/site manager total labour cost per employee and (2001) US manufacturing pay for skill/knowledge, profit/gain sharing, sales value/labour costs; same

plants; response rates meetings, and TQM (as index of involvement) respondent as for independent unclear variables. Data collected by US

Bureau of the Census19 Guthrie 164 New Zealand Single high-involvement work practices Questionnaire; single-source; Self-report: productivity (annual

(2001)c firms, heterogenous (HIWP) scale based on 12 practices various staff from CEO to junior sales per employee); same sample; 23 percent (e.g. performance-based promotion, manager respondent; strongly related to

skill-based pay, training, participation) objective data in a sub-sample;concurrent

20 Richard and 63–73 US banks; Single scale of effectiveness of strategic Questionnaire; single-source; Objective: return on equity (ROE) Johnson (2001) 13–15 percenta human resource management (SHRM) HR director and productivity (net income per

practices (from Huselid et al., 1997) (Effectiveness of HR practices, employee); concurrent tapping eight practices (e.g. participation not use per se)and empowerment, teamwork,employee and manager communications) Telephone survey; single- Self-report, percent change in sales

21 Batt (2002) 260 call centres; Skill level (education and training); job source; general managers in prior 2-year period; same 54 percent design (discretion and teamwork); HR respondent as for independent

incentives (supportive HR – training, variables; previous 2 years feedback, high pay, security). Composite HR index of all three

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 439

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Hu

man

Rela

tion

s 58

(4)

44

0

Table 1 continued


22 Way (2002) 446 small firms; Single index of ‘High Performance Telephone interview; single- Self-reportc labour productivity 70 percent Work Practices’ covering seven practices source; plant manager. Data (income/labour costs) plus

(e.g. self-directed teams, job rotation, collected by US Bureau of the perceived relative (to competitors) performance pay, training, involvement) Census productivity; concurrent

23 Ahmad and 107 manufacturing 12 practices including: selection, Self-report questionnaires; Self-report: single index covering – Schroeder plants in USA, teamwork, performance-related pay, respondents differ between cost, quality, delivery, flexibility, new (2003) Germany, Italy and training, employment security, practices, but single-source for product introduction; different

Japan; 60 percent harmonization any given measure; managers, respondent to that for the engineers, supervisors and independent variables; concurrentworkers

24 Guest et al. 366 manufacturing Single overall use of HRM practices (48 Telephone survey; single-source; Objective: productivity (sales per (2003) and service sector items on: recruitment and selection, training HR director or most senior employee) and profit for years

UK companies; and development, appraisal, financial flexibility, HR person before and after the period when response rate unclear job design, two-way communication, the independent variables were

employment security and internal measuredlabour market, harmonization, quality)

25 Wright et al. 50 business units, Single overall HR practices scale: nine items Employee attitude survey; Objective (from company records),(2003) US food services covering selection, pay for performance, multiple source; rank-and-file productivity and profit – subsequent,

company; response training, participation employees for period 3–9 months after rate unclear measurement of HR practices

a Estimated by current authors as the number of companies in relevant analyses as a percentage of total available sample.b It is a requirement to ensure the accuracy of all data supplied to the US Bureau of the Census.c See also Guthrie et al.’s (2002) article on the same data.

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 440

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


reduce error resulting from differences in the use of rating scales), raters whothrough knowledge of the use of practices in a range of organizations areable to benchmark the development of the practice against that elsewhere (tobetter calibrate the scale), and raters who are ignorant of the performanceor the organization in question (to eliminate the possibility that knowledgeof the outcome will contaminate measurement of the independent variables,resulting in common method variance bias). In contrast, the least satisfac-tory method is to use reports from single respondents, each describing theirown organization, and who have knowledge of, or worse still, are the sourceof, the performance data.

The profile of the large majority of studies corresponds to the leastsatisfactory measurement approach. Twenty-one of the 25 studies use singlerespondents (see Table 1, column 4). Moreover, in only two of the four caseswith multiple respondents (Wright et al., 2003; Youndt et al., 1996) is anyinformation on inter-rater agreement provided. In the other two cases(Ichniowski et al., 1997; MacDuffie, 1995), it is unclear how the informationwas combined, and the extent to which it was consistent across sources. Thisis not a trivial issue, given findings by Gerhart et al. (2000) and Wright etal. (2001) of very low levels of reliability in HR measures.

It is also the case that 24 of the 25 studies rely exclusively on respon-dents describing only their own context. The one exception is Ichniowski etal. (1997) who characterized the HRM system across the 36 US steelproduction lines in their sample themselves, on the basis of interview datafrom multiple respondents (though quite how is unclear). Moreover, in those24 studies it is reasonable to assume that those rating the extent of use ofthe HRM practices were also aware of their organization’s performance. Infact, this is certain in 12 cases in which data on HRM practices and perform-ance were obtained from the same respondent.

Thus, across the studies as a whole, measures of HRM are generallyof unknown reliability, are likely to contain considerable random measure-ment error, and at the same time in the majority of cases are open to conta-mination from knowledge of performance and demand characteristics.Random error will weaken the assessment of any underlying relationshipbetween HRM and performance; whereas contamination of the data willcreate correlated error that could result in spurious associations.

Dependent variable measurement

Measures of the dependent variable (performance) minimally should comefrom a different source from that used to measure HRM practices, andideally would be ‘objective’ – to reduce the likelihood of common method


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 441



variance. In many cases, such as in for-profit businesses, measures ofproductivity and profit taken from official accounts will be appropriate,perhaps supported by more industry-specific indicators (e.g. defects per1000 cars in auto assembly, as in MacDuffie, 1995). In other instancesdifferent measures will be appropriate, for example in charitable organiz-ations the percentage income redistributed will be relevant. So diversity ofperformance measures consistent with the nature of the business is to beexpected. Of course such measures are not necessarily totally accurate (e.g.organizations can bring forward or postpone costs for a given financialyear), but they are generally subject to external audit and, most importantly,are collected independently from the measurement of the HRM practices.This contrasts with the least desirable approach that relies on reportedperformance from a single employee who is also the source of informationabout the HRM practices.

The picture is more encouraging in this respect than for the measure-ment of HRM, but is still far from ideal (Table 1, column 5). Ten of the 25studies do use objective data, ensuring measures of performance come froma different source than those of HRM practices. The other side of the coin,of course, is that 15 of the studies rely on reported performance, in 12 ofwhich the information comes from the same person who assesses the HRpractices. It should be noted, however, that recent findings (Wall et al., 2004),suggest that such common source self-report performance data may notnecessarily be as biased as one might expect.

Research design

There are two main types of HRM-performance study. Cross-sectionalstudies are those in which both the independent and dependent variables aremeasured on one occasion only (be these concurrent or not); whereas inlongitudinal studies either or both the independent and dependent variablesare measured on at least two occasions. Both types of study are of value.Cross-sectional research is a cost-effective starting point for establishing thattwo or more variables are related, and the absence of a cross-sectionalrelationship would send warning signals that more costly longitudinal workmight not be justified. Also, cross-sectional work often allows the use ofmuch larger samples and hence augments generalizability. However, forreasons that are well known, cross-sectional studies provide a weakfoundation for causal inference, for which longitudinal research designs areto be preferred.

Within longitudinal studies, however, there are alternatives. Thesimplest advance on a cross-sectional design is what we shall call the


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 442



‘quasi-longitudinal’ study. Studies of this kind examine how the extent of useof HRM practices at one point in time predicts subsequent performance,while controlling for prior (or concurrent) performance. The advantage ofthis over a simple cross-sectional study (e.g. concurrent performance only) isthat it can help control for reverse causality, that prior performance fosteredgreater use of the HRM practices in question. This design, however, has someproblems, especially when the timing of the introduction of the HRM prac-tices is unknown. If those practices had been introduced, say, 5 years beforethe study, and had led to better performance over the 3 years prior to thestudy, then controlling for that prior performance would effectively beremoving the effect of interest. Consequently, whereas quasi-longitudinalresearch that shows a positive relationship with performance adds to ourunderstanding, if it fails to do so the interpretation is unclear.

A stronger option is what we shall call the ‘authentic longitudinal’study. Research of this kind would involve measurement of both the inde-pendent and dependent variables on two or preferably more occasions. Forexample, at Time 1 one would measure both HRM practices and perform-ance, at Time 2 the HRM practices again, and at Time 3 performance for asecond time. The timing of the measurements would be determined accord-ing to a hypothesized lag for the effect of HRM on performance. Evidenceconsistent with a causal interpretation would require change in HRM to beassociated with a subsequent change in performance. In this way, one istesting whether an increase (or decrease) in the use of the relevant practice(s)is associated with a subsequent improvement (or deterioration) in perform-ance; and stable third factors (e.g. sector, product mix) are controlled forthrough the repeated measures design. The design would be further improvedby focusing only on those organizations in which the HRM practices havebeen introduced (or substantially enhanced) between the performancemeasurement occasions; or by intervention studies (with controls).

Judged against the above criteria, the design of existing studies isdisappointing. As shown in Table 2, 21 of the 25 studies are purely cross-sectional (Table 2, column 2), providing very weak grounds for causal infer-ence. Moreover, 17 of these 21 cross-sectional studies are what Way (2002)describes as ‘temporally backward predictive’ (p. 779), that is to say theyinvolve a relationship between current HRM and prior performance. This isself-evident where authors explicitly report having used previous perform-ance data (e.g. Arthur, 1994; Batt, 2002). Moreover, even in those studiesthat use concurrent performance (e.g. Delery & Doty, 1996; Guthrie, 2001– or the average of previous and concurrent performance data as in the caseof Youndt et al., 1996), the grounds for inferring causality, were it to belinked to HRM, would be weak. This is because most concurrent measures,


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 443



Hu

man

Rela

tion

s 58

(4)

44

4

Table 2 Research design and findings of empirical studies on HRM and organizational performance

Study Research Control for Main effects (summary, Interaction Individual design third factors statistical significance, hit rate) effects practice effects

1 Arthur Cross-sectional Age, size, Commitment systems show higher Not examined Not examined(1994) unionization and reported productivity (p < .05) and

business strategy lower scrap rates (p < .05). 2/2 predicted effects

2 Guest and Cross-sectional No No differences across four types. Not examined Not examinedHoque (1994) 0/6 predicted effects

3 Huselid Cross-sectional For example size, Skills and structures related to GRATE Internal fit for skills and Not examined(1995) capital, R&D, intensity, (p < .05, one-tail); motivation related structures � motivation

unionization, sector to productivity and Tobin’s Q (p < .01, for GRATE; strategic fit one-tail). 3/6 predicted effects for motivation � HR

strategy – 2/14 4 MacDuffie Cross-sectional Degree of Work systems related to productivity Internal/organizational fit Not examined

(1995) automation, volume, (p < .05); and work systems and HRM – work systems � HRM complexity, product policies to quality (at p < .05 and p < .01, policies � buffers for design, age respectively). 3/4 predicted effects productivity; work systems

� buffers, for quality – 2/85 Delaney and Cross-sectional Many, e.g. size, HRM practices collectively (block in Selected internal fit Training and incentive

Huselid (1996) age, product vs. regression) associated with perceived interactions examined (e.g. pay associated with service, unionization, organizational and market performance staffing � training) but organizational sector (p values unspecified). 4/10 (i.e. five HR none found – 0/8 performance; staffing

practices � two performance measures) selectivity and predicted effects incentive with market

performance

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 444

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Wall &

Wood

Th

e ro

man

ce o

f HR

M4

45

6 Delery and Cross-sectional Size (assets), age, Appraisals (p < .05), profit sharing Strategic fit, individual HR As for main effectsDoty (1996) district (p < .01) and security (p < .01) related practices with

to ROA; only profit-sharing with ROE product/market (p < .01). 4/14 predicted effects innovation. One found

(appraisal � innovation) – 1/147 Koch and Cross-sectional Size, sector, HR planning and hiring related to HR sophistication � capital As for main effects

McGrath unionization, R&D, productivity (p < .05 and p < .01 intensity, showing the effect (1996) capital intensity respectively); development and of HR is stronger for higher

sophistication not. 2/4 predicted effects capital intensity – 1/18 Youndt et al. Cross-sectional Size, growth, change Administrative HR no effects; human Single-instance of strategic Not examined

(1996) in sales, product capital-enhancing HR related to fit – administrative HR �complexity, strategy reported productivity (p < .01), but quality strategy at p < .05 – (cost, quality, delivery, not to customer alignment or machine 1/8 scope) efficiency. 1/6 predicted effects

9 Huselid Quasi-longitudinal Unionization, size, Strategic HRM related to subsequent Not examined Not examinedet al. (1997) R&D, sector, GRATE (controlling for concurrent,

concurent p < .05) but not productivity or performance and Tobin’s Q; technical HRM unrelated others to any performance measure. 1/6

predicted effects10 Ichniowski Cross-sectional Up to 25 for technical Lines with innovative HRM systems No conventional test; but All components (e.g.

et al. (1997) and longitudinal reliability (e.g. age have greater productivity (p < .01); lines system dummies explain selection, training) and speed of line) becoming more innovative in HRM productivity over and above have individual effects,

over time showed increased additive effects of none additional to productivity (p < .01). 1/1 predicted component practices system results

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 445

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Hu

man

Rela

tion

s 58

(4)

44

6

Table 2 continued


11 Wood and Cross-sectional Many, e.g. size No difference across four types of Not examined Not examinedde Menezes sector, unionization workplace on productivity or change in (1998) productivity; some effect for financial

performance but not as expected. 0/3 predicted effects

12 Hoque (1999) Cross-sectional Size, unionization, Overall HRM related to productivity, Not formally tested, but Not examinedcountry of quality and financial performance sub-group analyses suggest ownership, strategy (p < .01 in all cases). 3/3 predicted effects interaction with business

strategy 13 Jayaram et al. Cross-sectional Size Cost (p < .01), flexibility (p < .05) and Not examined Not examined

(1999) time (p < .05). Each aspect of HRM related to its own outcome (e.g. HRM cost to cost performance);HRM generic related to time performance only (p < .05).4/8 predicted effects

14 Vandenberg, Cross-sectional No Flexibility positively associated with Not examined Not examinedet al. (1999) ROE; incentives and direction-sharing

negatively related to ROE; training and work design unrelated to ROE; p < levels unclear, and insufficient detail of analysis.1/5 predicted effects

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 446

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Wall &

Wood

Th

e ro

man

ce o

f HR

M4

47

15 Wright Cross-sectional Technology, size, age Only training related to financial Internal fit – selection, pay and As for main effectset al. (1999) performance – negatively (p < .05). 0/4 appraisal interact with

predicted effects participation (positive effect where participation is high,strong negative effect where participation is low) – 3/4

16 Bae and Cross-sectional Unionization, sector, HRM strategy positively associated Not examined Not examinedLawler (2000) country of origin with self-report (p < .05) and both

concurrent (p < .01) and subsequent (p < .05) objective (ROIC) performance measures. 3/3 predicted effects

17 Fey et al. Cross-sectional Sector, size, years Overal performance associated with Not examined As for main effects(2000) operating in Russia promotion on merit for managers, and

job security for other employees. 2/16 predicted effects

18 Cappelli and Longitudinal: two Many, including For none of the eight practices did their Internal fit between self- As for main effectsNeumark sets of panel production function, introduction relate to either productivity managed teams and profit (2001) data, 1977–93, age, sector or change in productivity in the sharing. No effects for nine

n = 433 and 1977–93 panel data set. Of the three other theoretically specified 1977–96, n = 666 practices in the 1977–96 panel data interactions (e.g. teamwork

set, one (job rotation) related to both training � self-managing outcomes, but negatively. Thus 0/8 and teams) – 1/10 0/3 predicted effects

19 Guthrie (2001) Cross-sectional Size, age, HIWP positively associated with HIWP with labour retention Not examinedunionization, sector, reported productivity (p < .01). 1/1 (positive effect of HIWP for relative pay predicted effect firms with high labour

retention, negative for low) – 1/1

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 447

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Hu

man

Rela

tion

s 58

(4)

44

8

Table 2 continued


20 Richard and Cross-sectional Size, part of holding SHRM not related to either ROE or Strategic fit, HRM with Not examinedJohnson (2001) company/not, productivity. 0/2 predicted effects capital intensity to predict

location, ROE (HRM has stronger organizational stage positive relationship with ROE (e.g. start up, growth) the greater the capital

intensity), but not productivity – 1/2

21 Batt (2002) Cross-sectional Sector, unionization, Work design and HR incentives Composite HR index interacts As for main effectscustomer segment positively related to sales growth with segment: stronger effect

(p < .05 and p < .01 respectively); no of HR on sales growth in effect for skill level. 2/3 predicted effects residential and smaller centres

22 Way (2002) Cross-sectional Sector, unionization, No statistically significant relationship Not examined Only, performance size, capital intensity with productivity ratio; positive pay associated with

relationship (p < .05) with overall perceived productivity perceived productivity (p < .05)

23 Ahmad and Cross-sectional Country, sector Practices related to performance at Not examined As for main effects,Schroeder p < .05 or p < .01 in 9/12 cases. But the but not controlling (2003) analysis fails to control for other for others

practices (all of which are highly intercorrelated), hence proportion of predicted effects unclear

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 448

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


Wall &

Wood

Th

e ro

man

ce o

f HR

M4

49

24 Guest et al. Quasi-longitudinal Size, sector, Use of HRM practices unrelated to Not examined Not examined(2003) unionization, subsequent productivity or profit (when

consultation, HR prior productivity or profit controlled).cost strategy, prior 0/2 predicted effectsperformance, others

25 Wright et al. Cross-sectional None, but all HR practices not statistically significantly Not examined Not examined(2003) (HR T1, business units in related to productivity but positively

performance T2 same corporation and related to profit (p < .05). 1/2 – 3–9 months fairly homogenous predicted effectslater)

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 449

©

2005 Th

e Tavisto

ck Institu

te. All rig

hts reserved

. No

t for co

mm

ercial use o

r un

auth

orized

distrib

utio

n.

at Middlesex U



Dow

nloaded from


and certainly financial ones, though ostensibly covering the same periodwhen the HR practices data were collected, necessarily reflect the result ofprior performance. There is a lag built into performance data, as it takes timeto sell, receive payment for and bank income, which is then aggregated overa financial year. At face value, therefore, such cross-sectional studies couldbe seen as examining the effect of performance on HRM practices ratherthan vice versa.

Of the four non-cross-sectional studies, two are quasi-longitudinal(Guest et al., 2003; Huselid et al., 1997). That is to say, they include repeatedmeasurement of performance, but a single measurement of HRM practices.The first of these studies controls for concurrent performance, and the otherfor prior performance (Table 2, column 5). However, in neither case is it clearwhether the earlier measurement of performance preceded the introductionof the HRM practices, so causal inference is difficult. Only two studies(Cappelli & Neumark, 2001; Ichniowski et al., 1997) have an authenticlongitudinal design on which causal inferences about the relationshipbetween HRM practices and performance could justifiably be based; and, aswe shall see later, they yield divergent results.

Control for third factors

Especially in cross-sectional studies there is a need to control for thirdfactors, that is, variables that may account for an association between HRMpractices and performance. For instance, if larger organizations, those withgreater (or lesser) trade union involvement, or those in particular sectors,tend to have both more sophisticated HR provisions and better performance,then a relationship between HR and performance would be found for oneor other of these reasons. Better studies will use controls for third factors,less satisfactory ones will not.

Overall, the studies do well in this respect. Control for third factors isevident in all but two studies (Table 2, column 3). Size is included in the largemajority of studies, but there is considerable variability with respect to whichother factors are controlled. Our judgement, however, is that the variabilityis largely consistent with the nature of the study (e.g. controls for sector inheterogenous samples, and for product complexity when examining build-hours on assembly lines).

Main effects

Our summary of findings for the relationship between HRM and perform-ance is based on the most sophisticated analysis presented by the authors,


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 450



but before any tests for interactions. Thus if both zero-order correlations andregression analyses controlling for third factors are presented, it is thefindings from this that are used. However, if regression analyses excludingand including interaction terms are presented, we rely on the findings fromthe former. We adopt the latter position to increase comparability betweenstudies that did and did not test for interactions (see next section). We alsoonly accept findings that are statistically significant at p < .05. For ease ofpresentation in Table 2 (columns 4–6), and in places in the text that follows,we refer to findings as effects in line with the allocation of variables to predic-tor and outcome status within regression analysis. However, we do notintend this to signify a causal link.

The main effect findings are summarized in Table 2, column 4. Theseshow that 19 of the 25 studies report some statistically significant positiverelationships between HRM practices and performance. At the same time,however, six studies fail to find any such relationships (e.g. Cappelli &Neumark, 2001; Guest et al., 2003), and in one of these (Wright et al., 1999)there is a statistically significant negative relationship (between training andfinancial performance). In parentheses, it is interesting to note that there isno evidence that studies using self-report measures of performance providemore positive findings than those using objective measures.

The balance of evidence towards positive relationships is weakenedby the lack of consistency within studies. The large majority used multipleperformance indicators, and many also used multiple HRM measures. Thismeans that there were multiple opportunities to show HRM–performancerelationships, of which only a few are statistically significant. This is shownby the hit rate also reported in Table 2, column 4. In Huselid’s (1995)seminal study, for example, we should judge the finding that the employeemotivation HRM scale was statistically significantly associated withproductivity and Tobin’s Q against the fact that the employee skills andstructures HRM scale was not statistically significantly related to either ofthese performance outcomes. Correspondingly, though the employee skillsand structures HRM scale was related to GRATE, the employee motivationHRM scale was not. Similarly, Delery and Doty (1996) identified sevenaspects of HRM and two performance indicators. Only 4 of the 14 relation-ships were statistically significant. Other studies involving more than asingle relationship are equally patchy (e.g. Youndt et al., 1996, with one ofsix predicted effects).

Moreover, effect sizes are typically small, and the criteria used to judgestatistical significance, and hence to draw conclusions about the reliability offindings, are often lenient, even in large sample studies. With a sample ofmore than 800, for example, Huselid (1995) recorded an effect size for his


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 451



two comprehensive measures of HRM combined on productivity of only 1percent, and treated one-tailed p-values of < .10 as statistically significant.Similarly, Wood and de Menezes (1998), also with a sample of over 800establishments, use a p < .10 statistical significance criterion.

The acceptance of small effect sizes and the use of moderate levels ofstatistical significance are perhaps indicative of a more general tendency inthe literature towards over-positive interpretations of results. Guest andHoque (1994), for instance, distinguished between four types of HRMsystem but found no statistically significant difference across them onreported productivity or quality. This did not prevent them concluding that‘the results demonstrate that strategic HRM pays off’ (p. 11). Virtually allauthors reach positive conclusions irrespective of the strength and consist-ency of their findings.

Interaction effects (fit)

All studies of the HRM–performance relationship would benefit from testsfor interaction effects, and this is so for several reasons. If the hypothesis isthat HRM practices have a universal effect on performance, then it addsstrength to any observed main effect to show it is invariant across differentcircumstances (e.g. according to sector or company size). Conversely, iftheory predicts internal, organizational or strategic fit, then tests of inter-actions are relevant for that purpose. Whatever the particular theoreticalrationale, the basic point is that investigating possible interaction effects is ameans of more fully understanding the nature of any observed relationshipbetween HRM practices and performance, and enhancing the constructvalidity of the study.

Twelve of the 25 studies do not report tests for interactions. Of theother 13, 2 test for interactions using non-standard methods (e.g. sub-groupanalysis), and the remaining 11 deploy orthodox tests (e.g. moderated regres-sion). Looking at those studies in more detail, the emphasis is on internaland strategic, rather than organizational, fit. However, the main impressionis of considerable variation in the interactions examined and inconsistencyin the findings recorded. Thus there is no compelling evidence either ofsynergy within HRM systems (i.e. that the effects of systems as a whole isgreater than the additive sum of their parts), or of any systematic strategicfit effects.

Individual practice effects

Most studies of the HRM–performance relationship aggregate individualpractices into multi-component scales (e.g. Huselid, 1995; Way, 2002), or


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 452



less commonly they group organizations on the basis of their HRM profile(e.g. Ichniowski et al., 1997; Wood & de Menezes, 1998), in order to testfor effects. An argument for this approach is that one would expect acomposite of HR practices to have a greater impact on overall organizationalperformance than any single practice. Nonetheless, all those compositemeasures include particular practices (e.g. training, empowerment, andperformance-related pay) each of which may enhance performance in its ownright. Thus the question arises as to whether one or a few among thoseconstituent practices may account more parsimoniously for any observedeffect of the overall HRM system on performance, or whether all are anintegral part of the whole. This is not only of theoretical importance, but isalso an issue with practical significance for managers wishing to stage theimplementation of beneficial HR practices and avoid ineffectual ones. Thusexamination of the relative effects of different component practices adds tothe construct validity of the investigation.

This issue has been ignored in the majority of studies. Fifteen of thestudies (Table 2, column 5) use composite measures of HRM without report-ing any findings for their component practices. Results for component prac-tices from the other 10 studies do not show any consistent pattern. Giventhis lack of evidence, it is difficult to be certain that it is the HRM systemthat relates to performance rather than one or particular sets of itscomponents.

A wider considerationA final methodological issue, not covered in Tables 1 and 2 because it islargely invariant, concerns the nature of the measurement of HRM practices.With only one exception (Huselid et al., 1997), studies have focused simplyon the use of the practices. It could be argued, however, that effective usemay be a better measure. This is because the inadequate implementation ofa given practice may do more damage than no implementation at all. Forexample, adopting team briefings without allowing employees to expressideas may have a worse effect than not considering such involvement at all.

Expert witness testimony: part 2

The second question addressed to our expert witness concerned whetheravailable evidence could be interpreted as establishing a causal link betweenHRM practices and performance. The advice has to be that the evidence ispromising but only circumstantial. The unknown reliability of measures ofHRM, the paucity of studies with adequate research designs, and the incon-sistent results both across and within studies is troublesome. Taken togetherwith the likelihood that positive findings are more likely to be published than


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 453



negative ones, it is clear that existing evidence for a relationship betweenHRM and performance should be treated with caution.

Summary and discussion

The conclusion from our analysis is that it is premature to assume that HRMinitiatives will inevitably result in performance gains, either in all situationsor even where deemed appropriate by contingency arguments. Consultantswho promote the HRM model are reflecting a wider academic and businessschool perspective in which this approach is central to definitions of moder-nity in management. Governments, employers’ associations, professionalassociations such as those representing personnel managers, and even tradeunions, throughout the world, have also promoted high-commitment HRMas the approach most suited to the assumed increasingly turbulent inter-national economy.

Our assessment is that, although consultants are acting in good faith,and their views are seemingly reinforced by the presumption on the part ofacademics that HRM systems actually do promote organizational perform-ance, the empirical evidence is as yet not strong enough to justify thatconclusion. The cross-sectional evidence could be over-estimating such arelationship due to contamination between measures of the HRM andperformance. Conversely, studies could be underestimating the strength andconsistency of the relationship through inadequate measurement of HRMpractices. We lean to the latter view, because while there is little evidence ofcommon method bias leading to spurious conclusions (Wall et al., 2004)there are strong indications of poor measurement of HRM practices (Gerhartet al., 2000) that would produce attenuated effects. Moreover, regardless ofthe quality of data, the paucity of longitudinal studies also makes causalinference dubious. So the evidence is at once encouraging but ambiguous.Because there also remain strong theoretical grounds for believing an HRMsystem centred on enhancing employee involvement should be beneficial fororganizational performance, research that overcomes the weaknesses ofcurrent studies is required.

Implications

There are three main implications of our analysis. First, there is a need, onthe basis of the findings so far, to temper the claims being made. A first stepis to moderate the language used. The term ‘high performance’ clearlypresupposes the very effects researchers should be investigating, and should


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 454



be avoided. Other terminological changes would also help foster a moretemperate debate. For example, use of words such as ‘impact’, ‘determinant’and ‘effect’ when presenting cross-sectional findings, a practice too oftenevident in the existing literature, should be avoided in favour of terms suchas ‘associated with’ or ‘related to’.

Second, adopting a strategy of articulating and investigating therelative merits of competing hypotheses may encourage more rigorous anddisinterested research. This has long been advocated (see Chamberlin’s‘method of multiple hypotheses’ as proposed in his presidential address tothe US Academy of Science in 1890; reprinted in Science, 1965) as a meansof combating bias in science, and is now more feasible than ever givendevelopments in computing and statistical methods, such as structuralequation modelling. By evaluating competing hypotheses it is less likely thatresearchers will align themselves with a particular hypothesis. One of theattractions of some of the early studies, most notably Huselid’s (1995) andDelery and Doty’s (1996), was precisely that they aimed to test between theuniversalistic and contingency theories of HRM. However, the selection ofcompeting hypotheses needs to be much broader. For example, these couldinclude hypotheses of differential impacts of HRM systems depending on theperformance measure used. Such development must, in our judgement, gohand-in-hand with theoretical development.

Third, and most importantly, the need is for research designs thatovercome the problems of existing studies. Among these, three weaknessesare paramount. The first is the reliance on single-source measures of HRMpractices (e.g. from CEO or HRM manager) of unknown reliability, sensi-tivity and validity, often from the same source as the measure of perform-ance. The second is the use of small samples coupled with low response rates.The third is the lack of sophisticated longitudinal studies, especially onesexamining how change in HRM practices relates to subsequent change inperformance. We now discuss the methodological implications of those threeweaknesses for future research.

Independent audit of HRM practices

Others have already pointed to the need to improve the reliability and validityof measures of HRM by obtaining data from multiple sources (e.g. Wright & Gardner, 2003). We would go further to suggest that it is essen-tial to have an independent audit of HRM and to cover comparison practices(e.g. R&D emphasis, total quality management). The audit should beconducted by individuals from outside the organization but familiar with thewider use of the practices in question, so that they can calibrate their


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 455



assessments accordingly. They would draw on multiple respondents andsources of data within organizations, and develop methods to achieve a highdegree of concordance across different auditors. They would measure boththe extent and effective use of practices, and be unaware of the performanceof the organization. The use of selected comparator practices would enabledetermination of the importance of HRM practices relative to others. Becausesize of effect can have limited meaning, it is instructive to know if HRM prac-tices have a stronger or weaker relationship with productivity than other prac-tices or factors (e.g. R&D expenditure). Expanding predictor variables in thisway also allows tests for possible interactions between HRM and such otherpractices (e.g. HRM with TQM, see Lawler et al., 1995).

Large samples, high response rates

Many of the important questions are ones that also require large samples.Testing for a relationship between HRM and performance benefits from thecontrol of third factors, and also from examination of the effect of individualcomponent practices relative to each other and to measures of overall HRM.The more such variables are taken into account, the more convincing anyrelationship is observed. However, the greater the number of variablesinvolved, the larger the sample size required.

Sample size is even more of an issue when addressing theoreticallypredicted interactions (i.e. the different types of fit), where the lack of sensi-tivity of orthodox moderated regression means large samples are essential(Aiken & West, 1991; Busmeyer & Jones, 1993). Of course, one could arguethat taking a universalistic perspective reduces the need for such large-scaleresearch. Yet, this is not so. Following our earlier call for testing competinghypotheses, it is important to see whether the effect of HRM on perform-ance is invariant across conditions (e.g. private vs. public sector; manufac-turing vs. finance; strategic fit) rather than leave that issue unaddressed.Finally, of course, large sample sizes, assuming also high response rates,provide a firmer base for generalization.

Longitudinal research designs

The final requirement we emphasize, as do others (Guest, 1997; Wright &Gardner, 2003), is the need for longitudinal research. Some such investi-gations could be retrospective. Thorough independent audit of HRM prac-tices, including evidence of when they were first introduced withinorganizations, could be set alongside available data on performance goingback many years. Then the effect of the introduction of practices on


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 456



performance could be estimated. Other studies should be prospective, basedon repeated independent audits, accepting the fact that findings will not beavailable for several years. The advantage of prospective over retrospectiveinvestigation is greater sensitivity and certainty in the measurement ofchange, and better coverage of both use and effective use of HRM practices.

The case for big science

The principles underlying our design criteria are well known and widelyaccepted as good practice. The weaknesses in the existing research designscannot, therefore, be attributed to investigators being unaware of how betterto conduct research. Rather, they stem from the almost inevitable limitationsof the small-scale funding that characterizes social science. Funding for anindividual or research team, over just a few years, does not provide theresources to deploy the recommended methods. Consider the need for inde-pendent audit, where experience suggests that it takes at least 2 days toobtain the requisite information for a single organization (see Patterson etal., 2004). Given such information should be obtained separately by at leasttwo individuals (to determine reliability), and for large samples of organiz-ations (say 400 in five key sectors), the data collection task alone would takemany person years (i.e. 2 people � 2 days � 2000 organizations = 8000days, which at 230 working days per year = some 35 person years). Add tothis the final recommendation, for the repeated measurement of practices andperformance as required by longitudinal research, and minimally the necess-ary resource is doubled.

Research on that scale is seemingly beyond the scope of current fundingmechanisms. It is not, however, beyond the scope of more intensive collabor-ation than is the current norm among the many potentially interested parties,including academics, employer organizations, employee organizations,professional bodies and government departments. The involvement ofgovernment agencies is also important to encourage the desirable level ofcompliance (i.e. high response rate). Moreover, there are precedents for suchactivity. There are already numerous information-collecting exercises inmany countries with which individuals (e.g. the census) and organizations(e.g. company financial reports) have to comply. There are assessmentswithin organizations based on independent audits (e.g. ‘Investors in People’in the UK); and there are surveys based on collecting data through interviewsthat are designed and managed by collaborative teams involving governmentdepartments, research councils and academics, as in the case of the WorkEmployee Relations Survey (WERS) series in the UK, or the equivalentAustralian survey.


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 457



We propose that a big science project on management practices andperformance builds on the best features of these exercises by devising a newstudy that covers practices well beyond employment relations and HRM,uses independent audit of those practices, and is at the level of the companyor firm (rather than site) for which there is official financial performancedata. If academics, practitioners and policy makers are serious in wishing tounderstand the effect of HRM and other practices on performance, thenthese kinds of data will be required. They must also face the possibility, ofcourse, that no strong causal effects will be found.

Postscript

It will be appreciated that the focus of this article is on a particular issueconcerning HRM that has dominated recent research – the relationship ofhigh involvement forms of HRM with organizational performance. Thisshould not be allowed to mask the fact that, even if such a performance effectwere not demonstrable, the use of practices that enhance employee involve-ment may be an end in itself. Such personnel management practices also maybe an important means of providing justice and equality of opportunity.Other reasons for studying HRM practices and systems could then include,for example, their role in enhancing employee well-being and promotingnon-discriminatory practices. It will also be recognized that the large-samplequantitative approach taken here is a reflection of the question addressed.Where an HRM effect on performance is established, then important ques-tions concerning the conditions under which this applies, the mechanismsthrough which it operates, and any unintended consequences of the HRMpractices, will require supplementary quantitative and qualitative investi-gations. Such detailed research can also shed considerable light on why noperformance effects accrue when that is found to be the case. In other words,our recommendation for big science should not be pursued to the exclusionof complementary options.

Acknowledgements

The support of the Economic and Social Research Council (ESRC) UK is grate-fully acknowledged. The work was part of the programme of the ESRC Centrefor Organisation and Innovation.


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 458



References

Ahmad, S. & Schroeder, R.G. The impact of human resource management practices onoperational performance: Recognising country and industry differences. Journal ofOperations Management, 2003, 21, 19–34.

Aiken, L.S. & West, S.G. Testing and interpreting interactions. Newbury Park, CA: Sage,1991.

Appelbaum, E., Bailey, T., Berg, P. & Kalleberg, A.L. Manufacturing advantage: Why highperformance work systems pay off. Ithaca, NY: Cornell University Press, 2000.

Arthur, J.B. Effects of human resource systems on manufacturing performance andturnover. Academy of Management Journal, 1994, 7, 670–87.

Bae, J. & Lawler, J.J. Organizational and HRM strategies in Korea: Impact on firmperformance in an emerging economy. Academy of Management Journal, 2000, 43,502–17.

Barney, J. Firm resources and sustained competitive advantage. Journal of Management,1991, 17, 99–120.

Barney, J. Looking inside for competitive advantage. Academy of Management Executive,1995, 9, 49–61.

Batt, R. Managing customer services: Human resource practices, quit rates, and salesgrowth. Academy of Management Journal, 2002, 45, 587–97.

Becker, B.E. & Gerhart, B. The impact of human resource management on organizationalperformance: Progress and prospects. Academy of Management Journal, 1996, 39,779–801.

Becker, B.E. & Huselid, M.A. High performance work systems and firm performance: Asynthesis of research and managerial implications. In K.M. Rowlands & G.R. Ferris(Eds), Research in Personnel and Human Resource Management. Greenwich, CT: JAIPress, 1998.

Busmeyer, J. & Jones, L.R. Analysis of multiplicative causal rules when causal variablesare measured with error. Psychological Bulletin, 1993, 93, 549–62.

Cappelli, P. & Neumark, D. Do ‘high performance’ work practices improve establishment-level outcomes? Industrial and Labor Relations Review, 2001, 54, 737–75.

Chamberlin, T.C. The method of multiple working hypotheses. Science, 1890, 15.(Reprinted in Science, 1965, 148, 754–9.)

Cooper, C. In for the count. People Management, 2000, October, 28–34.Delaney, J.T. & Godard, J. An IR perspective on the high-performance paradigm. Human

Resource Management Review, 2001, 40, 472–89.Delaney, J.T. & Huselid, M.A. The impact of human resource management practices on

perceptions of organizational performance. Academy of Management Journal, 1996,39, 919–69.

Delery, J.E. Issues of fit in strategic human resource management: Implications for research.Human Resource Management Review, 1998, 37, 289–309.

Delery, J.E. & Doty, D.H. Modes of theorizing in strategic human resource management:Tests of universalistic, contingency, and configurational performance predictions.Academy of Management Journal, 1996, 39, 802–35.

Dyer, L. & Reeves, T. Human resource strategy and firm performance: What do we knowand where do we need to go? International Journal of Human Resource Management,1995, 6, 656–70.

Fey, C.F., Björkman, I. & Pavlovskaya, A. The effect of human resource management prac-tices on firm performance in Russia. International Journal of Human Resource Manage-ment, 2000, 11, 1–18.

Gerhart, B., Wright, P.M., McMahan, G.C. & Snell, S.A. Measurement error in researchon human resources and firm performance: How much error and does it influence effectsize estimates? Personnel Psychology, 2000, 53, 803–34.


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 459



Godard, J. A critical assessment of the high-performance paradigm. British Journal ofIndustrial Relations, 2004, 42, 349–78.

Guest, D. Human resource management and performance: A review and research agenda.International Human Resource Management, 1997, 8, 263–76.

Guest, D. & Hoque, K. The good, the bad and the ugly: Employee relations in new non-union workplaces. Human Resource Management Journal, 1994, 5, 1–14.

Guest, D.E., Michie, J., Conway, N. & Shehan, M. Human resource management andcorporate performance in the UK. British Journal of Industrial Relations, 2003, 41,291–314.

Guthrie, J.P. High involvement work practices, turnover and productivity: Evidence fromNew Zealand. Academy of Management Journal, 2001, 44, 180–90.

Guthrie, J.P., Spell, C.S. & Nyamori, R.O. Correlates and consequences of high involve-ment work practices. International Journal of Human Resource Management, 2002,13, 183–97.

Hoque, K. Human resource management and performance in the UK hotel industry. BritishJournal of Industrial Relations, 1999, 37, 419–43.

Huselid, M.A. The impact of human resource management practices on turnover, produc-tivity, and corporate financial performance. Academy of Management Journal, 1995,38, 635–72.

Huselid, M.A., Jackson, S.E. & Schuler, R.S. Technical and strategic human resourcemanagement effectiveness as a determinant of firm performance. Academy of Manage-ment Journal, 1997, 40, 171–88.

Ichniowski, C. Human resource management systems and the performance of US manu-facturing businesses. Working paper No. 3349. Cambridge, MA: National Bureau ofEconomic Research, 1990.

Ichniowski, C., Shaw, K. & Prennushi, G. The effects of human resource management prac-tices on productivity. American Economic Review, 1997, 87, 291–313.

Jayaram, J., Droge, G. & Vickery, S.K. The impact of human resource management prac-tices on manufacturing performance. Journal of Operations Management, 1999, 18,1–20.

Kalleberg, A.L. & Moody, J.W. Human resource management and organizational perform-ance. In A.L. Kalleberg, D. Knoke, P.V. Marsden & J.L. Spaeth (Eds), Organizations inAmerica: Analyzing their structures and human resource practices. Thousand Oaks, CA:Sage, 1996.

Koch, M.J. & McGrath, R.G. Improving labor productivity: Human resource managementpolicies do matter. Strategic Management Journal, 1996, 17, 335–54.

Lawler, E.E. High involvement management. San Francisco: Jossey-Bass, 1986.Lawler, E.E., Mohrman, S.A. & Ledford, G.E., Jr. Creating high performance organiz-

ations. San Francisco: Jossey-Bass, 1995.Lawler, E.E., Mohrman, S.A. & Ledford, G.E., Jr. Strategies for high performance organiz-

ations. San Francisco: Jossey-Bass, 1998.MacDuffie, J.P. Human resource bundles and manufacturing performance: Organizational

logic and flexible production systems in the world auto industry. Industrial and LaborRelations Review, 1995, 48, 197–221.

Marchington, M. & Grugulis, I. Best practice human resource management: Perfect oppor-tunity or dangerous illusion? International Journal of Human Resource Management,2000, 11, 905–25.

Miles, R.E. & Snow, C.C. Designing strategic human resource systems. OrganizationalDynamics, 1984, 13, 36–52.

Osterman, P. Work reorganization in an era of restructuring: Trends in diffusion and effectson employee welfare. Industrial and Labor Relations Review, 2000, 53, 179–96.

Patterson, M.G., West, M.A. & Wall, T.D. Integrated manufacturing, empowerment andcompany performance. Journal of Organizational Behavior, 2004, 25, 641–65.


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 460



Pfeffer, J. Competitive advantage through people: Unleashing the power of the workforce.Boston: Harvard University Press, 1994.

Philpott, J. Productivity and people management. London: Chartered Institute for Person-nel Development, Perspective Series, Spring, 2002.

Richard, O.C. & Johnson, N.B. Strategic human resource management effectiveness andfirm performance. International Journal of Human Resource Management, 2001, 12,299–310.

Schuler, R.S. & Jackson, S.E. Linking competitive strategies with HRM management strate-gies. Academy of Management Executive, 1987, 1, 207–19.

Vandenberg, R.J., Richardson, H.A. & Eastman, L.J. The impact of high involvement workprocesses on organizational effectiveness. Groups and Organization Management,1999, 24, 300–99.

Wall, T.D., Michie, J., Patterson, M., Wood, S.J., Sheehan, M., Clegg, C.W. & West, M.A.On the validity of subjective measures of company financial performance. PersonnelPsychology, 2004, 57, 95–118.

Walton, R.E. From ‘control’ to ‘commitment’ in the workplace. Harvard Business Review,1985, 63, 77–84.

Way, S.A. High performance work systems and intermediate indicators of firm perform-ance within the US small business sector. Journal of Management, 2002, 28, 765–85.

Womack, J., Jones, D.T. & Roos, D. The machine that changed the world. New York:Rawson, 1990.

Wood, S.J. Human resource management and performance. International Journal ofManagement Reviews, 1999, 1, 367–413.

Wood, S.J. & de Menezes, L. High commitment management in the UK: Evidence fromthe Workplace Industrial Relations Survey and Employers’ Manpower and Skills Prac-tices Survey. Human Relations, 1998, 51, 485–515.

Wood, S.J. & Wall, T.D. Human resource management and business performance. In P.B.Warr (Ed.), Psychology at work. Harmonsworth: Penguin, 2002.

Wright, P.M. & Gardner, T.M. The human resource–firm performance relationship:Methodological and theoretical challenges. In D. Holman, T.D. Wall, C.W. Clegg, P.Sparrow & A. Howard (Eds), The new workplace: A guide to the human impact ofmodern working practices. Chichester: Wiley, 2003.

Wright, P.M., Gardner, T.M. & Moynihan, L.M. The impact of HR practices on theperformance of business units. Human Resource Management Journal, 2003, 13,21–36.

Wright, P.M., Gardner, T.M., Moynihan, L.M., Park, H.J., Gerhart, B. & Delery, J.E.Measurement error in research on human resources and firm performance: Additionaldata and suggestions for future research. Personnel Psychology, 2001, 54, 875–901.

Wright, P.M., McCormick, B., Sherman, W.S. & McMahan, G.C. The role of humanresource practices in petro-chemical refinery performance. International Journal ofHuman Resource Management, 1999, 10, 551–71.

Youndt, M.A., Snell, S.A., Dean, J.E. & Lepak, D.P. Human resource management, manu-facturing strategy, and firm performance. Academy of Management Journal, 1996, 39,836–65.


03 055032 Wall (to_d) 20/6/05 1:05 pm Page 461




Toby Wall is Professor of Work Psychology at the University ofSheffield, where he is Director of the Institute of Work Psychology andof the ESRC Centre for Organization and Innovation. His interests spanwork design, innovation, employee well-being, absence, human resourcemanagement, work and organizational performance, and applied researchdesign. He has published 12 books and over 150 articles in books andjournals. The latter include papers in the Academy of Management Journal,Journal of Applied Psychology and Personnel Psychology. He has providedpolicy advice to government departments and to numerous organiz-ations.[E-mail: [email protected]]

Stephen Wood is Research Chair and Deputy Director, Institute ofWork Psychology, Professor of Employment Relations, School of Manage-ment, and Co-Director, ESRC Centre for Innovation and Organisation,University of Sheffield. His current interests include high involvementmanagement, lean production, work design, idea capturing schemes,family-friendly management, employment relations and the social andeconomic challenges of nanotechnology. Publications include (with H.Gospel) Representing workers: Trade union recognition and membership inBritain (Routledge, 2003); The transformation of work (Unwin Allen, 1989);and (with Richard Jones and Alison Geldart), The social and economic chal-lenges of nanotechnology (Economic and Social Science Research Council,2003). He has provided policy advice to government departments,professional associations and public and private sector organizations.[E-mail: [email protected]]

03 055032 Wall (to_d) 20/6/05 1:05 pm Page 462



Education

Romance Of Hr And Performance